Consider adding DarcyWeisbachResistance as a new node type

[gijsber]

What
Consider a mixture of physical representations for different flow conditions. For suggestions see: discussion
#2446

Why
The current Manning implementation is slow when head differences are tiny and conditions are laminar instead of turbulent

[Huite]

I think my current recommendation would be to replace it by a DarcyWeisbachResistance (or something) instead, since it's derived on more fundamental grounds rather than empiric Manning or Chezy.

However, many users will be much more familiar with Manning parametrizations and the like, so a constructor/utility to convert to equivalent roughness values may be worthwhile (albeit less transparent perhaps...).

[simulutions]

I propose as first research task to base the root relaxation we do now on the reynolds number. If this gives us a lot of numerical stability and performance gain we can then transition to a more physical based linearization in the laminar regime.

This research task can probably be performed in a morning to see if it actually is worth while to continue.

[Huite]

Do you mean starting the relaxation at Re < 3000 (approx. laminar flow) instead of the current arbitrary threshold of 1e-5?

Quick note, if you check the wikipedia Moody diagram (which is admittedly for pipes, not open channels),

the friction factor is not monotonic with the Reynolds number, which is a bit unfortunate for the Newton-Raphson solve. But if you paste over the transition zone with a nice sigmoid, it'll probably work. In the worst case, you can also paste over the transition zone with a flat line and keep it monotonic...

[simulutions]

"starting the relaxation at Re < 2000 (approx. laminar flow) instead of the current arbitrary threshold of 1e-5?" is exactly what I mean and then calculate the Reynolds number for open channel flow:

[simulutions]

My proposal of relaxing based on Re < 2000 makes the model incredibly slow.
I have benchmarked the model HollandseDelta_parameterized_2025_7_1, which runs in the old implementation in 6 minutes, 29 seconds, 154 milliseconds.

If I turn on the relaxation based on the reynolds number, julia estimates the ETA on the order of:

From these results it does not seem that manning still really is slow..